221 results about "Expectation–maximization algorithm" patented technology

The invention provides a large model patent atlas construction method with constraints. The method comprises the steps that a TRIZ rule table is constructed, TRIZ rules of each process are matched by inputting atlas states and atlas construction processes, the TRIZ rules of the processes form a process TRIZ rule table, TRIZ rule matching is carried out according to each process, and TRIZ rule matching results are obtained; converting the updated constraint condition cluster into a logical reasoning rule, performing constraint condition cluster classification on knowledge in the graph, judging whether the graph knowledge violates constraint conditions or a dependency set according to the logical reasoning rule, and if the graph knowledge violates the constraint conditions or the dependency set, triggering early warning; and carrying out correlation analysis on the updated constraint condition cluster by adopting an expectation maximization algorithm, iteratively optimizing the constraint condition cluster until the correlation of the constraint condition cluster tends to be stable, and updating a logical reasoning rule and an early warning mode according to the optimized constraint condition cluster.

A method for estimating channel parameters of a reconfigurable intelligent surface based on a spherical wave assumption includes the following steps. In Step 1, a signal transmission model of a RIS-assisted near-field communication is constructed based on the spherical wave assumption; in Step 2, channel measurement data in different RIS transmission modes are obtained; in Step 3, a delay, an angle of arrival, an angle of departure, a Doppler shift and a polarization matrix of multipath in channels are estimated based on a space-alternating generalized expectation maximization algorithm, and angle parameters, distance parameters and coupling polarization matrices of the multipath at a RIS end are estimated based on a maximum likelihood principle; and in Step 4, the estimated parameters are updated and iterated subsequently. The method can estimate all important channel parameters in the RIS-assisted near-field communication scenario more accurately.

The invention belongs to the technical field of blast furnace system prediction and health management, and particularly discloses a blast furnace system residual life prediction method based on an adaptive calibration model. The method comprises the following steps: firstly, establishing an initial degradation model by utilizing partial observable blast furnace wall temperature data; estimating hyper-parameters in the degradation model based on maximum likelihood estimation, and designing a trigger mechanism based on a dynamic threshold to judge whether to perform adaptive calibration on parameters or structures of the degradation model; when the model is not triggered, updating the model parameters by adopting a Kalman filter; when the triggering condition is met, the model structure is calibrated based on the prediction error, parameters of the prediction error model are identified through a strong tracking filter and an expectation maximization algorithm, and the optimal structure of the prediction error model is determined through an akaike information criterion. According to the method, the parameters and the structure of the degradation model can be adaptively calibrated based on online data, the prediction precision of the residual life of the blast furnace system can be improved, and the online adaptability is high.

The invention discloses a non-reference channel passive radar networking cooperative weak target signal level positioning method, and the method comprises the steps: firstly initializing system parameters, and building a target echo model, which is polluted by noise and direct wave interference, of a non-reference channel passive radar; then, a complete data set is established in combination with a distributed multi-station target echo model and an unknown signal of a non-cooperative external radiation source, a conditional likelihood function of the complete data set is derived, and finally, based on an expectation maximization principle, an expectation step and a maximum step of the conditional likelihood function are executed in a loop iteration mode so as to continuously update target position estimation and suppress direct wave interference. And the maximum likelihood estimation of the target position is obtained until the convergence condition of the expectation maximization algorithm is met, and combined direct wave suppression and target position estimation updating are realized. The method provided by the invention solves the problems that the current passive radar system is low in positioning precision, easy to be interfered by direct waves, unknown in non-cooperative external radiation source signal, strong in dependence on a reference channel and the like in weak target detection.

The invention discloses a line intelligent planning and fault early warning method suitable for electric power engineering design, and relates to the technical field of electric power engineering. The method comprises the following steps: quantifying geographic information, geological conditions and construction cost into a continuous decision curved surface through spatial semantic segmentation and a cost curved surface model, and identifying a large-area low-cost region as a primary feasible region in combination with a dynamic clustering algorithm; a multi-dimensional decision index is constructed based on spatial features, geological stability and facility association data, fault regions such as terrain abrupt change or facility conflict are identified through real-time association coefficients, and an exception handling mechanism is triggered; and selecting a seed region with an optimal construction condition as an anchor point, constructing a joint probability model of spatial features and cost, and generating an optimal path considering feasibility probability and cost effectiveness in combination with an expectation maximization algorithm and a graph optimization technology. According to the invention, scientificity and economical efficiency of cable planning under complex geological conditions are improved, and key technical support is provided for intelligentization of electric power engineering.

The invention discloses a flexible touch screen surface defect automatic detection method and system, and the method comprises the following steps: S1, compensating the illumination of a surface image of a flexible touch screen through a Retinex method, and converting the illumination to an HSV color space, and obtaining a converted image; s2, calculating a directional feature map based on the converted image; s3, constructing a Gaussian mixture model based on the directional feature map, and clustering the features by using an expectation maximization algorithm to obtain an initial defect region; s4, optimizing the initial defect area by utilizing morphological processing to obtain an optimized defect area; and S5, splicing the converted image and the optimized defect area to obtain an input image, and performing defect detection on the input image based on a deep learning-based multi-scale feature fusion network. According to the invention, rapid and accurate detection of the surface defects of the flexible touch screen can be realized.

The invention discloses a mechanical structure reliability evaluation method based on a hybrid R-Vine Copula model, and the method comprises the steps: firstly, selecting an optimal R-Vine tree type according to sample data by employing an AIC criterion and the basic definition of an R-Vine tree type, employing a hybrid Pair-Copula function for the first layer of R-Vine tree type to improve the sample adaptability, employing Clayton, Gumbel and Frank Copula functions to construct a hybrid Copula function, and employing a Pair-Copula function to construct a Copula model; calculating a weight coefficient and a dependent parameter of the mixed Pair-Copula function by using an expectation maximization algorithm; a subsequent tree type is constructed by adopting a single Pair-Copula function, so that the optimal balance between the calculation efficiency and the calculation precision is achieved, and the optimal construction based on the mixed R-Vine Copula model is completed; and finally, solving the reliability index and the failure probability of the structure based on a mixed R-Vine Copula model in combination with an improved first-order second-order moment method.

The invention discloses a lithium battery residual life prediction method based on a multi-stage Wiener process, and relates to the technical field of lithium batteries, and the method comprises the steps: building a lithium battery degradation model based on the Wiener process; self-adaptively estimating a degradation state and model parameters based on an expectation maximization (EM) algorithm and a Kalman smoothing method; and in order to predict the residual life at the current moment, considering the condition that the degradation state at the change point is unknown, solving a probability density function of the residual life, and obtaining an estimated value of the residual life. According to the lithium battery residual life prediction method based on the multi-stage Wiener process, the degradation model based on the multi-stage Wiener process with the measurement error is established, the probability density function of the residual life is solved and simplified, and the accuracy of lithium battery residual life prediction is improved.

The invention discloses a bad weather degraded image restoration method and device based on dynamic degradation generation, and relates to the field of image processing, and the method comprises the steps: employing an E step of an expectation maximization algorithm to optimize a hidden variable in a training process of a noise predictor and a dynamic degradation generator, and obtaining an optimized hidden variable; in M steps of an expectation maximization algorithm, inputting the optimized hidden variables and state variables into a dynamic degradation generator, generating a degradation layer to perform semi-supervised training on a noise predictor and the dynamic degradation generator, and obtaining a trained noise predictor; in an image restoration stage, an image degraded in severe weather and random Gaussian noise are input into a trained noise predictor, an iterative denoising step obeying a Markov chain is executed, and a clean restored image is obtained. The problems that an existing severe weather image restoration model is limited in generalization ability and unstable in restoration effect are solved.

The invention discloses a bearing residual life prediction method and device and a medium, and relates to the field of bearing life prediction, and the method comprises the steps: carrying out the feature extraction of a vibration signal time sequence of a target bearing, and obtaining a time domain feature, performing time domain feature screening, feature weight assignment and weighted calculation based on robustness, monotonicity and predictability to obtain a standby feature vector; performing multiple regression operation on the to-be-used time domain feature set and the to-be-used feature vector set by adopting a partial least square method to realize feature fusion to obtain a comprehensive degradation feature quantity; and on the basis of the comprehensive degradation characteristic quantity corresponding to each vibration signal time sequence, constructing a prediction model by adopting a preset index model, performing parameter updating and estimation on the prediction model by adopting a Bayesian updating and expectation maximization algorithm, and performing prediction error correction by adopting a strong tracking filtering algorithm so as to obtain a bearing residual life prediction result. The method can improve the prediction precision of the residual life of the bearing.

The invention discloses a model training method and device, a storage medium and a computer program product, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining a weight positive sample training set and a weight negative sample training set, the weight positive sample training set and the weight negative sample training set are obtained by labeling and screening a preset unlabeled sample set; based on a preset unbiased positive sample and unmarked sample learning algorithm, according to the weight positive sample training set and the weight negative sample training set, training to obtain a dichotomy model; iterating the dichotomy model through a loss function and a regular function corresponding to the dichotomy model based on a preset expectation maximization algorithm until the dichotomy model converges, and obtaining a converged dichotomy model; and taking the converged dichotomy model as a prediction model. The problems that an existing method is poor in universality and generalization and needs to label samples manually are solved, and the universality and generalization of model training are improved.

The invention relates to the technical field of target tracking, and discloses a maneuvering target tracking system and method based on hybrid coding and UKF decoding. The maneuvering target tracking method comprises the steps of obtaining radar measurement data of a target position at each moment based on a target trajectory; the method comprises the following steps: encoding radar measurement data by adopting an attention-GRU network to obtain a target state encoding value; based on the target state coding value, obtaining a corrected target state noise covariance matrix, a corrected target state measurement noise covariance matrix and a corrected state initial covariance matrix by adopting an expectation maximization algorithm; and performing UKF decoding calculation on radar measurement data by adopting the corrected target state noise covariance matrix, the corrected target state measurement noise covariance matrix and the corrected state initial covariance matrix to obtain a target position. Model parameters such as noise variance information do not need to be preset, the tracking precision of the maneuvering target can be effectively improved, and noise interference is weakened.

The invention discloses a multi-dimensional distribution offset adaptive threshold detection method and system, and the method comprises the steps: obtaining real-time data, extracting a multi-dimensional statistical feature, and obtaining a feature vector; based on historical normal data, using an expectation maximization algorithm to train a Gaussian mixture model, and determining parameters to obtain a normal distribution model; inputting the feature vector into the model, and calculating a probability value of the feature vector belonging to normal distribution as a first offset judgment index; based on the real-time data distribution of a plurality of detection objects in the same group, the distribution difference of any two objects is calculated by using a Wasserstein distance, and the similarity between the objects is obtained; and constructing a similarity network and calculating connectivity as a second offset judgment index. Setting a fixed-length sliding window, dynamically updating two indexes in the window, and obtaining a first self-adaptive threshold value and a second self-adaptive threshold value; and when any index is lower than a corresponding threshold value, determining distribution offset and giving an alarm, and updating model parameters in real time by using an incremental expectation maximization algorithm. According to the invention, accurate detection and intelligent analysis of data distribution offset are realized.

The invention provides a communication auxiliary sensing system and method based on an intelligent super surface (RIS). The communication auxiliary sensing system comprises a multi-antenna base station, a multi-user terminal and an active RIS. According to the system, an uplink pilot frequency and a data signal are jointly utilized, an expectation maximization (EM) algorithm is adopted to iteratively recover equivalent channel state information (CSI), and a Cramer-Rao bound (CRB) of sensing performance is deduced. Combined optimization of user precoding and RIS reflection coefficients is innovatively proposed under the uncertainty of CSI, so as to minimize CRB and guarantee user communication rate at the same time. An original problem is converted into a processable form by establishing a traversal MIMO spectrum efficiency compact lower bound, and efficient optimization is realized by combining a WMMSE method with a block coordinate descent (BCD) frame. According to the scheme, the sensing precision and the spectrum efficiency are remarkably improved, the method is suitable for an active RIS-assisted MU-MIMO network, and the technology can also be expanded to a traditional RIS-free system.

The invention provides a short-distance high-speed optical fiber communication equalization method and system, and the method comprises the steps: building a Gaussian mixture model for an optical signal subjected to the quadrature amplitude modulation of a set order, and completing the parameter initialization of the Gaussian mixture model through a particle swarm optimization algorithm of which the inertia weight is firstly increased and then decreased; the method comprises the following steps: inputting data features of two dimensions of an in-phase component and an orthogonal component of data into a Gaussian mixture model, then solving parameters of the Gaussian mixture model by using an expectation maximization algorithm, and determining the membership degree of each data point and a clustering label corresponding to each unary Gaussian distribution through the parameters; and through the maximum posterior probability, the data are distributed to the corresponding clustering labels, and nonlinear equalization is completed. According to the method, the memory property of signal nonlinear impairment can be accurately captured, the nonlinear clustering precision is improved, and the calculation complexity of a nonlinear clustering model is reduced under the same clustering effect.

The invention provides a vulnerability repair method based on a dynamic denoising and hybrid enhancement technology, which comprises the following steps: S1, performing dynamic denoising and hybrid enhancement on an original vulnerability repair model to obtain an optimized vulnerability repair model; s11, deriving approximate distribution of clean data and noise data from the training loss of the original vulnerability repair model by using an expectation maximization algorithm; s12, calculating the weight of the clean data and the weight of the noise data by using a confidence coefficient function of distribution perception, and re-weighting the approximate distribution by using the weight of the clean data and the weight of the noise data; s13, generating a mixed sample by using the interpolation coefficient and reconstructing a loss function to obtain a total loss value; s14, repeatedly training the vulnerability repair model by using the value of the total loss until the value of the current total loss is not reduced in continuous n rounds or reaches the total number of rounds of training, and outputting the optimized vulnerability repair model; and S2, performing vulnerability repair by using the optimized vulnerability repair model. The software development efficiency and quality are improved, and software vulnerabilities are reduced.

The invention discloses a deception jamming signal identification and interference source positioning method and system, and belongs to the technical field of signal identification and interference source positioning, and the method comprises the steps: receiving intermediate frequency receiving signals of a global satellite navigation system, including a real satellite signal, a forwarding deception jamming signal, a multipath signal and a noise signal; performing multi-component parameter estimation based on a space alternating generalized expectation maximization algorithm to obtain a parameter set comprising time delay, Doppler frequency shift and complex amplitude; according to consistency detection of receiver clock differences and Doppler frequency shift information of a plurality of adjacent base stations, identifying deception jamming components and real signal components from the parameter set; and determining the position of an interference source by adopting a delay inequality positioning algorithm based on the arrival delay inequality of the deception interference component among the plurality of adjacent base stations. According to the method, the GNSS signal characteristics are utilized, space-power time-frequency joint detection is realized with low complexity under the condition of not needing extra hardware cost, and the performance of deception jamming detection in practical application is improved.

The invention discloses a fiber laminate damage acoustic emission signal processing method, equipment and a medium, and relates to the technical field of signal processing.The method comprises the steps that a fiber laminate is fixed to a test device, signals are collected through an acoustic emission sensor, filtering and preprocessing are carried out, features are extracted through an improved ReliefF algorithm, and a fiber laminate damage acoustic emission signal is obtained; randomly selecting samples, comparing the samples with nearest neighbors, updating feature weights, and screening most relevant features. An expectation maximization algorithm and a Gaussian mixture model are combined to establish a damage identification model, different types of damage are automatically classified, and a Gaussian distribution model is obtained through iterative optimization. And by utilizing a clustering analysis and feature selection algorithm, the damage of the laminated plate with different layering angles is identified, and different damage types are accurately distinguished. According to the method, high-precision damage classification is realized through feature extraction in combination with an expectation maximization algorithm and a Gaussian mixture model, laminated plate damage analysis at different layering angles is dynamically adapted, and feature extraction efficiency, classification accuracy and model adaptability are remarkably improved.

The invention belongs to the technical field of wireless communication, and particularly relates to a positioning method based on low-orbit satellite opportunity signal Doppler measurement. The method solves the Doppler measurement positioning problem based on low orbit (LEO) satellite opportunity signals under the condition that GNSS signals are limited or unavailable. A positioning process is modeled as a non-linear state space framework, a joint estimation scheme combining a cubic Kalman smoother and an expectation maximization algorithm is provided, efficient estimation of a terminal position and Doppler frequency shift is realized, and noise covariance is adaptively optimized at the same time. The method provided by the invention theoretically ensures the optimal performance in statistical significance. A simulation result shows that compared with a traditional nonlinear least square method, the scheme provided by the invention can remarkably improve the positioning precision, and the calculation overhead is kept reasonable and balanced.

The invention relates to the technical field of voice recognition, and discloses a voiceprint model training method, a user identity detection method and device, and an intelligent household electrical appliance, and the voiceprint model training method comprises the steps: carrying out the preprocessing of a voice signal, and extracting a mixed feature vector containing a static frequency spectrum feature and a time difference feature; training a Gaussian mixture model to be converged by adopting an expectation maximization algorithm based on the mixed feature vector; and outputting the special voiceprint model of the speaker. The voiceprint model obtained by training through the training method is used for user identity detection, and high robustness, high real-time performance and high accuracy of voiceprint recognition in a complex noise environment are achieved.

The invention discloses a natural electric field-electric field joint detection and inversion method based on GMM prior, and relates to the technical field of tunnel engineering geological detection, and the method comprises the steps: obtaining the resistivity and natural potential observed in a to-be-inverted region; an equivalent body source of a natural potential is parameterized into a body source vector on a grid unit, weighted L1 sparse regular constraint is introduced into the body source vector, a GMM prior model is used as underground electrical structure prior shared across physics fields, and a multi-task joint inversion objective function of a shared underground electrical structure label is established; and performing inversion iteration by adopting maximum posteriori, updating parameters of the GMM prior model and the underground electrical structure label by adopting an expectation maximization algorithm until an end condition is met, and obtaining a joint inversion imaging result of the tunnel water-rich anomalous body. The advantages of a multi-source electrical method are fused, underground electrical structure priori is introduced, Bayesian uncertainty analysis is combined, and high-resolution and high-reliability recognition of the water-rich anomalous body in front of the tunnel is achieved.

The invention discloses an electronic device residual life online prediction method and system under an operation condition, a storage medium and electronic equipment, and the method comprises the steps: building a Gamma state space model, and carrying out the processing through an expectation maximization algorithm, and obtaining an initial parameter estimation value of the Gamma state space model; and based on the initial parameter estimation value and pre-acquired real-time monitoring data of the field operation device, dynamically updating the real health state of the electronic device through a kernel smoothing particle filter algorithm, and estimating new parameter estimation of the Gamma state space model by using an expectation maximization algorithm to obtain a final Gamma state space model. According to the method, the degradation information is extracted and the initial degradation model is established based on the accelerated failure experiment data, and then the degradation information and the model parameters are dynamically updated by adopting the kernel smoothing particle filter algorithm in combination with the field operation data, so that the real-time life prediction under the field operation condition is realized.

The invention provides an earthquake landslide disaster chain risk assessment method, medium and equipment, and relates to the technical field of geological disaster chain risk assessment, and the method comprises the steps: obtaining earthquake-induced disaster chain data, and constructing an earthquake landslide disaster chain knowledge graph; according to the knowledge graph, acquiring disaster influence factors, determining association among the disaster influence factors, and determining key disaster influence factors with relatively high association degree; key disaster influence factors are used as Bayesian network nodes, node variable data are discretized, a maximum-minimum hill climbing algorithm is used for structural learning, and a directed acyclic graph structure of the Bayesian network is determined; performing Bayesian network parameter learning by using an expectation maximization algorithm to obtain a prior probability of Bayesian network nodes; if the prediction result of the model does not reach the expectation, the association between the nodes is adjusted, and if the prediction result reaches the expectation, the model is used for earthquake landslide disaster chain risk assessment. According to the scheme, the reliability of the prediction result can be improved.

The invention discloses a Bayesian network-based rainstorm-mountain torrent disaster chain trigger threshold identification method, which comprises the following steps: according to a preset fuzzy character string matching algorithm and a preset text similarity algorithm, carrying out cleaning processing on extracted historical disaster situation texts in sequence to obtain processed disaster situation texts; performing feature extraction on the processed disaster text to obtain key disaster features; constructing a target Bayesian network topology of the rainstorm-mountain torrent disaster chain; discretizing each node variable in the target Bayesian network topology to obtain a state value of the node variable; the prior probability of the node variables is calculated, and the conditional probability of the causal relationship between the nodes is calculated according to the prior probability and a preset application expectation maximization algorithm; calculating a target posterior probability that the first disaster triggers the derived disaster; and according to a preset posterior probability critical value and the target posterior probability, inverting the intensity of the first disaster, and determining an intensity threshold value of the derived disaster triggered by the first disaster.

The invention discloses a personalized picture book intelligent recommendation system based on combination of innovative sentence vectors and a hidden semi-Markov model, which belongs to the technical field of education and comprises a data acquisition module, a sentence vector generation module, an interest tracking module, a recommendation generation module, the innovative sentence vectors, the hidden semi-Markov model and an expectation maximization (EM) algorithm. According to the method, an efficient and accurate personalized picture book recommendation system is successfully constructed through deep fusion of the innovative sentence vectors and the HSMM, the problem of interest drifting of students in the reading process is effectively solved, the innovative sentence vectors can comprehensively capture deep semantics and emotional tendencies of reading texts of the students, and the interestingness of the students is improved. A rich and multi-dimensional data basis is provided for interest analysis, the HSMM model allows the students to stay for different time durations in different potential interest states, the heterogeneity of the interest duration of the students is accurately described, and dynamic tracking of interest changes of the students is achieved.

The invention relates to the technical field of wind tunnel flow sound separation, in particular to a method for separating acoustics and related pneumatic wall pressure fluctuation in a closed wind tunnel, which comprises the following steps: acquiring sound pressure of a measurement signal in the closed wind tunnel, and storing the sound pressure in a cross-spectrum matrix; constructing a related pneumatic wall pressure fluctuation interference model of the closed wind tunnel according to physical factors such as convection velocity in the closed wind tunnel, a related transverse attenuation degree and a related longitudinal attenuation degree; constructing a related pneumatic wall pressure fluctuation interference model and a related pneumatic wall pressure fluctuation interference-acoustic separation model of probability factor analysis according to the low-rank property and the sparsity; according to the sparse prior information, constructing a prior model to constrain a related pneumatic wall pressure fluctuation interference-acoustic separation model; and obtaining maximum likelihood estimation of the parameters through an expectation maximization algorithm, and determining contribution degrees of acoustics and related pneumatic wall pressure fluctuation pressure of the to-be-tested component in the closed wind tunnel. The problem that an existing method does not consider interference physical information and is difficult to be directly applied to closed wind tunnel flow-sound separation is solved.

The invention discloses a method and a system for measuring a time average velocity field of a combustion chamber of a scramjet engine, and belongs to the technical field of average velocity field measurement. The method comprises the following steps: establishing a measurement system, matching the number of exposure frames of a high-speed camera with the number of cavities of a laser, collecting a time sequence image for pre-processing, obtaining the pre-processed time sequence image, dividing the time sequence image into grid slices for cross-correlation calculation, and obtaining a time sequence speed according to a space and pixel conversion coefficient; constructing a velocity model according to a Gaussian mixture model algorithm, and obtaining a probability matrix of a data point belonging to a correct velocity vector according to an expectation maximization algorithm; and calculating to obtain a time average velocity field according to the correct velocity vector and the error velocity vector judgment condition. The method effectively improves the efficiency and accuracy of measuring the distribution of the time average velocity field, and can be used for measuring the time average velocity field.

The invention relates to the technical field of satellite navigation positioning, in particular to a Beidou robust Kalman filtering positioning method based on a non-Gaussian noise model. According to the method, non-Gaussian noise generated by shielding and a multipath effect in Beidou positioning is accurately represented by constructing a Gaussian mixture model and a student t distribution model, model parameters are updated online by an expectation maximization algorithm, and the problem that a traditional Gaussian hypothesis model cannot adapt to noise statistical characteristics is solved; the non-Gaussian noise model is fused with robust filtering, a self-adaptive robust weight matrix is constructed through a Huber cost function, an IGGIII scheme or noise probability density, an observation covariance matrix is dynamically adjusted, the influence of abnormal observation values is suppressed, the defects that classical robust filtering lacks noise modeling and weight adjustment fails under continuous strong interference are overcome, and the robustness of the robust filtering is improved. And meanwhile, the problem of complex particle filtering calculation is avoided.

The invention discloses an Internet of Things anomaly detection method and system based on quaternion state space diffusion enhancement, and belongs to the technical field of network security and artificial intelligence. The method comprises the following steps: mapping a flow time sequence feature into a quaternion tensor to maintain an internal coupling relationship of a multi-dimensional feature; a double-flow encoder is designed, a quaternion selective state space model is adopted to extract continuous fluid features, and a dynamic hypergraph neural network is adopted to model discrete protocol features; carrying out self-supervised pre-training on a resistance pseudo-anomaly sample by utilizing potential diffusion model generation, and optimizing characterization by combining quaternion cepstrum distance loss; the injected learnable prompt vector is optimized in the small sample fine tuning stage, and a category prototype is corrected by using a semi-supervised expectation maximization algorithm; and calculating a sample anomaly score based on an energy model to realize known attack classification and unknown anomaly judgment. According to the method, the generalization ability of the model under the small sample condition and the unknown threat detection ability are remarkably improved.

The present invention relates to the technical field of data communication, and relates to a precise time synchronization clock parameter estimation method based on optimal invariant estimation. In the present invention, in a communication scene where unknown random time delay distribution information is present, on the basis of a precise time protocol, timestamp interaction is achieved by means of bidirectional data packet transfer between a master node and a slave node, and a clock relation model between the master node and the slave node is established; considering that a plurality of intermediate devices exist in the data packet interaction process, resulting that random time delay distribution information cannot be accurately obtained, a Gaussian mixture model is used to model a random time delay distribution, a competitive expectation maximization algorithm is used to learn mixture model parameters; and finally, on the basis of the parameters obtained by learning, an optimal invariant estimation method is used to estimate a clock frequency offset and a clock phase offset simultaneously, so as to realize time synchronization between the master node and the slave node. The present invention improves the accuracy of clock frequency offset estimation and clock phase offset estimation, is suitable for a communication scene where unknown random time delay distribution information is present, and enhances the robustness of the synchronization algorithm to a communication time delay.